Workers aged 22 to 25 in the most AI-exposed occupations have seen employment fall roughly 13% relative to their peers in less-exposed jobs since late 2022, according to Stanford's Digital Economy Lab. Among young software developers specifically, employment is down nearly 20% from its late-2022 peak through mid-2025. Workers over 30 in the same occupations show no comparable decline. That is not a prediction about the future of work. That is what the payroll data already shows.

This article is grounded in current advisory work, not retrospective analysis. Mark Lynd is a 5x CEO/CIO/CISO with Thinkers360 Top 10 global rankings across Cybersecurity and Artificial Intelligence and was ranked #1 globally in Cybersecurity in 2023. He is currently Head of Executive Advisory and Strategy at Netsync, advising enterprise C-Suites and boards on the AI and cybersecurity questions moving fastest in 2026. The frameworks and patterns referenced here are from active engagements this quarter.

What The Data Actually Show

Researchers Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen analyzed payroll records covering millions of US workers and published their findings, titled "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," in August 2025. Their headline finding is not the dramatic, economy-wide job loss that dominates the AI-and-jobs conversation. It is narrower and, for that reason, more credible. There is no evidence of widespread displacement across the workforce. What the data show instead is concentrated in one specific slice, entry-level workers in occupations that are both AI-exposed and easy to automate rather than merely AI-adjacent.

Two details in the research matter more than the headline number. First, pay for the young workers who kept their jobs did not drop. The effect shows up in hiring and headcount, not in wages for people already employed. That distinguishes this from a story about employers cutting compensation. It is a story about who gets hired in the first place. Second, the researchers split AI use into two categories, automative and augmentative. Automative use is where the software executes the task that a person used to do, drafting the routine code, answering the routine ticket, generating the routine document. Augmentative use is where AI assists a person who remains the one doing the work. Employment fell in roles dominated by automative use, in programming and accounting-adjacent entry roles. Employment grew in roles dominated by augmentative use, even among young workers in the same broad industries.

The occupational contrast in the underlying data is worth naming directly, because it undercuts the simplest version of the AI-and-jobs story. Young workers in customer service and software development roles, occupations where large language models can now execute a large share of the routine task directly, show the sharpest relative declines. Young workers in occupations like health aide roles, where the work is physical, relational, and hard for a model to execute end to end, show employment holding steady or rising over the same period. That is not a technology story about which industries adopted AI fastest. Health care organizations have adopted AI tools aggressively too. It is a story about which specific tasks inside a role a model can currently execute outright versus which tasks still require a person physically or socially present to do them.

The Augmentation Line

That distinction is the concept that matters more than any industry label. Call it the augmentation line, the divide running through every occupation, sometimes through every job description, between tasks where AI executes the output and tasks where a human still executes it with AI's help. It is not a line between tech jobs and non-tech jobs. It is not a line between AI-using companies and companies that avoid AI. It runs inside individual roles, and it is the strongest predictor the current data offers for where entry-level hiring is contracting versus holding steady.

A software company hiring two kinds of junior staff illustrates the line cleanly. One team hires junior engineers to write and review boilerplate code, the exact task category where AI-assisted code generation has matured fastest and where a senior engineer with an AI coding tool can now cover work that used to require two or three juniors. That team's junior hiring plan shrinks, not because leadership decided to replace people with software, but because the marginal junior hire no longer clears the bar the way it did three years ago. A second team hires junior sales development reps who use AI to draft outreach emails and summarize call notes, but who still make the calls, build the relationships, and close the deals themselves. That team's headcount plan is untouched, because the AI tool sits inside a workflow a human still owns end to end. Same company, same AI budget line, two completely different hiring trajectories, and the difference is exactly where each role sits relative to the augmentation line.

That has a direct consequence for how a company should be reading its own headcount data. A hiring freeze on junior engineers can look, from the outside, indistinguishable from a hiring freeze on junior sales reps. Both show up as the same line on the same spreadsheet. But one reflects a structural shift in how much junior labor a senior employee with AI tools can now cover, and the other reflects an ordinary budget decision unrelated to any of it. Leadership that does not disaggregate hiring data by where each role sits on the augmentation line will draw the wrong conclusion from both kinds of freeze, either overreacting to normal budget tightening or underreacting to a role category that is genuinely shrinking for structural reasons.

The Strongest Case Against This Reading

The fair objection to building a future-of-work argument on this study is real, and the study's own authors raise a version of it. Young workers, and tech-adjacent young workers specifically, were also hit hardest by the 2022 and 2023 interest rate increases and the tech sector layoffs that followed, independent of anything to do with AI. Software development hiring fell across the board in that period for reasons that had nothing to do with generative AI maturity, including venture funding pullbacks and a broader correction after 2021's hiring surge. A skeptic can reasonably argue that AI exposure is correlated with, rather than causal to, the occupations that got hit, because the same occupations that are AI-exposed also happen to be the ones most sensitive to tech-sector capital cycles.

The researchers address this directly rather than glossing over it, and that is part of why the study holds up. They report that the employment gap persists after excluding tech-sector firms specifically, after controlling for remote-work status, and after accounting for interest-rate movements, which rules out the simplest versions of the confound. They are more candid, however, that the effect weakens somewhat once education level is controlled for, and they describe their own findings as "canaries in the coal mine," early descriptive signals rather than definitive proof of causation. That caveat deserves to be taken at face value rather than argued away. The honest position is that this is the best current evidence of a real, AI-linked hiring effect concentrated in a specific population, not proof of a broader transformation of work, and not yet a closed case.

What Leadership Should Ask Monday Morning

For leadership and the board, the useful move is treating entry-level hiring as a place to look for early signal, not a place to assume the answer.

Which roles in our organization sit on the automative side of the augmentation line, where AI increasingly executes the task rather than assisting the person doing it.

Has our entry-level hiring plan for those roles changed in the last eighteen months, and can we say honestly whether that was a deliberate decision or a drift nobody named out loud.

Where AI use is augmentative, are we still building the pipeline of junior talent who will need to become the senior talent making judgment calls a decade from now.

If entry-level hiring keeps contracting in automative roles, who trains the next generation of people qualified to supervise the AI doing that work.

When we look at our own hiring data by role and by age cohort, does the pattern match what Stanford found, or does our organization look different, and can we explain why.

The data are honest about their own limits, and leadership should be too. What we know is that something real is happening to young workers in specific roles. What we do not yet know is how far the line runs, and pretending otherwise, in either direction, is the actual risk.